This application for an NIMH Mentored Research Scientist Career Development (K01) award entitled "Measurement error in latent class models of adherence" seeks support to provide an intensive, mentored research experience, culminating in a successful investigator-initiated research award application and launching a career as an academic biostatistician focusing on developing statistical methods for difficult methodological problems affecting the validity of mental health services research. As part of the education plan, the PI will increase her understanding of mental disorders and the systems used to collect services research data and explore a new area of statistical methods (latent class and latent variable methods). The four-phase research plan focuses on measurement error, latent class and latent variable methods for research questions generated from the NIMH study of "Prevention of Suicide in Primary Care Elderly: Collaborative Trial," but apply to data from randomized trials in general. The conventional "measurement models" under the SEM framework do not naturally accommodate validation data from a sub-sample. The classical and Berkson measurement models that appear in the statistical literature, offer different ways of incorporating validation data as a measurement error variance estimate like a variance component. The goals of the K and my future research focuses on incorporating these different measurement error approaches. Phase 1 is designed to understand how patterns of adherence are related to depression trajectories and how the tow influence treatment effects. This phase will focus on generating knowledge of latent class and latent variable methodology with application to the PROSPECT study. Phase 2 focuses on estimating measurement error in the measures of adherence. Phase 3 uses results from Phases 1 and 2 to adjust latent class/variable results for error in the measurement of adherence. Phase 4 will use everything learned in Phase 1, 2, and 3 to guide the development of a NIH R01 for statistical methods that accurately accommodate measurement error. In addition, Bayesian model averaging is proposed to account for model error by eliminating the choice of the number of latent classes. This research plan will provide useful tools that will aid in the design and analysis of other mental health services studies. [unreadable] [unreadable] [unreadable]